习题11.1
写出图11.3中无向图描述的概率图模型的因子分解式。
解答
P
(
Y
1
,
Y
2
,
Y
3
,
Y
4
)
=
1
Z
ψ
c
1
(
Y
1
,
Y
2
,
Y
3
)
ψ
c
2
(
Y
2
,
Y
3
,
Y
4
)
P(Y_1,Y_2,Y_3,Y_4)=\frac{1}{Z} \psi_{c_1}(Y_1,Y_2,Y_3) \psi_{c_2}(Y_2,Y_3,Y_4)
P(Y1,Y2,Y3,Y4)=Z1ψc1(Y1,Y2,Y3)ψc2(Y2,Y3,Y4)
Z
=
∑
Y
ψ
c
1
(
Y
1
,
Y
2
,
Y
3
)
ψ
c
2
(
Y
2
,
Y
3
,
Y
4
)
Z=\sum_Y\psi_{c_1}(Y_1,Y_2,Y_3) \psi_{c_2}(Y_2,Y_3,Y_4)
Z=∑Yψc1(Y1,Y2,Y3)ψc2(Y2,Y3,Y4)
习题 11.2
证明
Z
(
x
)
=
α
n
T
(
x
)
⋅
1
=
1
T
⋅
β
0
(
x
)
Z(x)=\alpha_{n}^{\mathrm{T}}(x) \cdot 1=1^{\mathrm{T}} \cdot \beta_{0}(x)
Z(x)=αnT(x)⋅1=1T⋅β0(x)，其中1是元素均为1的m维列向量。
解答
本式子出现在书的199页11.3.1前向-后向算法这一小节
PS：书中是
β
1
(
x
)
\beta_1(x)
β1(x)，但是我个人觉得这儿应该是
β
0
(
x
)
\beta_0(x)
β0(x)
Z
(
x
)
=
(
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
+
1
(
x
)
)
s
t
a
r
t
,
s
t
o
p
Z(x)=(M_1(x)M_2(x)...M_{n+1}(x))_{start,stop}
Z(x)=(M1(x)M2(x)...Mn+1(x))start,stop即
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
+
1
(
x
)
M_1(x)M_2(x)...M_{n+1}(x)
M1(x)M2(x)...Mn+1(x)得到的结果矩阵
M
M
M(mm维度)的
(
s
t
a
r
t
,
s
t
o
p
)
(start,stop)
(start,stop)位置的元素。其中
M
n
+
1
(
x
)
M_{n+1}(x)
Mn+1(x)为m*m维度的矩阵，但是只有stop列为1，其余为0。或者说
Z
(
x
)
Z(x)
Z(x)的值为
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
(
x
)
M_1(x)M_2(x)...M_{n}(x)
M1(x)M2(x)...Mn(x)结果矩阵
M
′
M'
M′的start行的所有元素之和。
α
n
T
(
x
)
⋅
1
=
α
n
−
1
T
(
x
)
M
n
(
x
)
⋅
1
=
α
n
−
2
T
(
x
)
M
n
−
1
(
x
)
M
n
(
x
)
⋅
1
=
.
.
.
=
α
0
T
(
x
)
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
(
x
)
⋅
1
=
Z
(
x
)
\alpha_{n}^{\mathrm{T}}(x) \cdot 1\\=\alpha_{n-1}^{\mathrm{T}}(x)M_n(x) \cdot 1\\=\alpha_{n-2}^{\mathrm{T}}(x)M_{n-1}(x)M_n(x) \cdot 1\\=...\\=\alpha^T_0(x) M_1(x)M_2(x)...M_n(x)\cdot 1\\=Z(x)
αnT(x)⋅1=αn−1T(x)Mn(x)⋅1=αn−2T(x)Mn−1(x)Mn(x)⋅1=...=α0T(x)M1(x)M2(x)...Mn(x)⋅1=Z(x)
说明：
α
0
T
(
x
)
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
(
x
)
\alpha^T_0(x) M_1(x)M_2(x)...M_n(x)
α0T(x)M1(x)M2(x)...Mn(x)得到的是1m维度的行向量，其值为
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
(
x
)
M_1(x)M_2(x)...M_{n}(x)
M1(x)M2(x)...Mn(x)的
s
t
a
r
t
start
start行的元素，将其与1是元素均为1的m维列向量点乘，得到的即为
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
+
1
(
x
)
M_1(x)M_2(x)...M_{n+1}(x)
M1(x)M2(x)...Mn+1(x)的
(
s
t
a
r
t
,
s
t
o
p
)
(start,stop)
(start,stop)位置的元素值。
1
T
⋅
β
0
(
x
)
=
1
T
⋅
M
1
(
x
)
β
1
(
x
)
=
.
.
.
=
1
T
⋅
M
1
(
x
)
M
2
(
x
)
M
3
(
x
)
.
.
.
M
n
(
x
)
⋅
β
n
+
1
(
x
)
=
Z
(
x
)
1^{\mathrm{T}} \cdot \beta_{0}(x)\\=1^T\cdot M_1(x)\beta_1(x)\\=...\\=1^T\cdot M_1(x)M_2(x)M_3(x)...M_n(x)\cdot \beta_{n+1}(x)\\=Z(x)
1T⋅β0(x)=1T⋅M1(x)β1(x)=...=1T⋅M1(x)M2(x)M3(x)...Mn(x)⋅βn+1(x)=Z(x)
说明：同理，
M
1
(
x
)
M
2
(
x
)
M
3
(
x
)
.
.
.
M
n
(
x
)
⋅
β
n
+
1
(
x
)
M_1(x)M_2(x)M_3(x)...M_n(x)\cdot \beta_{n+1}(x)
M1(x)M2(x)M3(x)...Mn(x)⋅βn+1(x)得到的是列向量，其每个值为
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
(
x
)
M_1(x)M_2(x)...M_{n}(x)
M1(x)M2(x)...Mn(x)的对应的一行元素之和（除了start列处，其余元素为0，与
1
T
1^T
1T点乘后得到的即为
M
1
(
x
)
M
2
(
x
)
.
.
.
M
n
+
1
(
x
)
M_1(x)M_2(x)...M_{n+1}(x)
M1(x)M2(x)...Mn+1(x)的
(
s
t
a
r
t
,
s
t
o
p
)
(start,stop)
(start,stop)位置的元素值。
习题11.3
写出条件随机场模型学习的梯度下降法.
参考Blog
习题11.4
参考图11.6的状态路径图，假设随机矩阵
M
1
(
x
)
,
M
2
(
x
)
,
M
3
(
x
)
,
M
4
(
x
)
M_1(x),M_2(x),M_3(x),M_4(x)
M1(x),M2(x),M3(x),M4(x)分别是
M
1
(
x
)
=
[
0
0
0.5
0.5
]
,
M
2
(
x
)
=
[
0.3
0.7
0.7
0.3
]
M_{1}(x)=\left[\begin{array}{cc}0 & 0 \\ 0.5 & 0.5\end{array}\right], \quad M_{2}(x)=\left[\begin{array}{cc}0.3 & 0.7 \\ 0.7 & 0.3\end{array}\right]
M1(x)=[00.500.5],M2(x)=[0.30.70.70.3]
M
3
(
x
)
=
[
0.5
0.5
0.6
0.4
]
,
M
4
(
x
)
=
[
0
1
0
1
]
M_{3}(x)=\left[\begin{array}{cc}0.5 & 0.5 \\ 0.6 & 0.4\end{array}\right], \quad M_{4}(x)=\left[\begin{array}{cc}0 & 1 \\ 0 & 1\end{array}\right]
M3(x)=[0.50.60.50.4],M4(x)=[0011]
求以start=2为起点stop=2为终点的所有路径的状态序列y的概率及概率最大的状态序列.
解答
y
=
(
1
,
1
,
1
)
=
a
21
b
11
c
11
=
0.5
∗
0.3
∗
0.5
=
0.075
y=(1,1,1)=a_{21}b_{11}c_{11}=0.5*0.3*0.5=0.075
y=(1,1,1)=a21b11c11=0.5∗0.3∗0.5=0.075
y
=
(
1
,
1
,
2
)
=
a
21
b
11
c
12
=
0.5
∗
0.3
∗
0.5
=
0.075
y=(1,1,2)=a_{21}b_{11}c_{12}=0.5*0.3*0.5=0.075
y=(1,1,2)=a21b11c12=0.5∗0.3∗0.5=0.075
y
=
(
1
,
2
,
1
)
=
a
21
b
12
c
21
=
0.5
∗
0.7
∗
0.6
=
0.21
y=(1,2,1)=a_{21}b_{12}c_{21}=0.5*0.7*0.6=0.21
y=(1,2,1)=a21b12c21=0.5∗0.7∗0.6=0.21（最大）
y
=
(
1
,
2
,
2
)
=
a
21
b
12
c
22
=
0.5
∗
0.7
∗
0.4
=
0.14
y=(1,2,2)=a_{21}b_{12}c_{22}=0.5*0.7*0.4=0.14
y=(1,2,2)=a21b12c22=0.5∗0.7∗0.4=0.14
y
=
(
2
,
1
,
1
)
=
a
22
b
21
c
11
=
0.5
∗
0.7
∗
0.5
=
0.175
y=(2,1,1)=a_{22}b_{21}c_{11}=0.5*0.7*0.5=0.175
y=(2,1,1)=a22b21c11=0.5∗0.7∗0.5=0.175
y
=
(
2
,
1
,
2
)
=
a
22
b
21
c
12
=
0.5
∗
0.7
∗
0.5
=
0.175
y=(2,1,2)=a_{22}b_{21}c_{12}=0.5*0.7*0.5=0.175
y=(2,1,2)=a22b21c12=0.5∗0.7∗0.5=0.175
y
=
(
2
,
2
,
1
)
=
a
22
b
22
c
21
=
0.5
∗
0.3
∗
0.6
=
0.09
y=(2,2,1)=a_{22}b_{22}c_{21}=0.5*0.3*0.6=0.09
y=(2,2,1)=a22b22c21=0.5∗0.3∗0.6=0.09
y
=
(
2
,
2
,
2
)
=
a
22
b
22
c
22
=
0.5
∗
0.3
∗
0.4
=
0.06
y=(2,2,2)=a_{22}b_{22}c_{22}=0.5*0.3*0.4=0.06
y=(2,2,2)=a22b22c22=0.5∗0.3∗0.4=0.06
扫一扫
624
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
